With the introduction of coding agents in more workflows, it’s interesting seeing where the bottlenecks are starting to reform. So far I see two: code reviews, and deployments. Naturally more code per unit time means more code to review. It seems like more and more of my day are just responding to these requests for reviews. Uninteresting work, but still quite necessary, I suppose.

This was expected, but what I didn’t expect was the pressure placed on the size of deployments. More services are being touched, since doing so is cheaper. In my personal experience, this is mainly because using such agents make it easier to do a more “proper” solution. There have been times in the past where I bodge a crappy fix in one service because I didn’t want to touch five others. This is less of a concern now, but it does mean that deployment will involve six services instead of just one.

We’ll see how this changes in the future. I have head some interesting things around how code review tools are adapting to larger and more frequent reviews, such as adapting are more semantic approach to showing what changed, rather than the half-century old diffs¹ we’re using now. As for micro-services, I still feel they’re becoming a relic of the manual coding age and that monoliths may be more useful in the future. Or maybe I’m just wish-casting here.